A neural net for your tip-of-the-tongue moments

What it does

WantWords runs the dictionary backwards: type a description, get back words that fit the meaning. It is designed to rescue you from the tip-of-the-tongue problem, help language learners, and even assist anomia patients who can describe an object but cannot name it. The backend is a Django service wrapping a dedicated neural reverse-dictionary model.

The interesting bit

Most NLP demos chase benchmarks; this one chases forgotten vocabulary. The authors published the core Multi-channel Reverse Dictionary Model at AAAI and the full system at EMNLP, explicitly targeting neurological word-retrieval failure as a use case.

Key highlights

• Semantic retrieval: returns words matching a natural-language definition rather than simple keyword inversion.
• Built on the MultiRD model, with pre-trained weights and training data available for direct download.
• Ships as a live web service (wantwords.net) and a mobile MiniProgram.
• Open-sources the full research stack: code, models, data, and citation-ready papers.
• Dependency manifest is a time capsule: Django 2.2.5, Torch 1.2.0, and pytorch-transformers 1.2.0.

Caveats

• The requirements list pins specific older releases (Django 2.2.5, Torch 1.2.0, pytorch-transformers 1.2.0), so expect compatibility work on a modern stack.
• The README details the backend pipeline and model but offers almost no information about the frontend JavaScript implementation.

Verdict

Grab it if you are researching semantic retrieval, building educational language tools, or want an NLP project with a concrete, non-benchmark application. Pass if you need a dependency-fresh, frontend-first modern stack.